This talk reports on analyses of service system data from a moderate size telephone call center. The data covers one year of operation and contains records from over 1,000,000 calls. About 35% of these calls involve a customer who desires to reach a call-center agent. These are the focus of our current study.
The event history of these calls can be divided into three primitive interacting, components: the arrival process, the queue (for those calls that need to queue), and the service process. We show the arrivals can be very well modeled as a time-inhomogeneous Poisson process. We then examine the service time of this process. We show that the service times are well modeled by a lognormal distribution with parameters depending on the available covariates. We also examine the behavior of customers in the queue with respect to the outcomes of customer service or abandonment.
We conclude with some calculations and observations about the relationship of our empirical results to those available from standard queueing theory models, and point to the necessity of developing and using certain modifications of that theory.
|